Orange Collective
Sazabi

Sazabi

The AI-native observability platform for fast-moving engineering teams.

sazabi.comYYC Profile@sazabiLinkedIn
Sazabi Product Overview[5]

Seed

Round

Unannounced · 100+ angels

9

Team size

Founded 2025

$14.2B

Observability TAM

By 2028 · Gartner forecast

Backed by

Harrison Chase

CEO, LangChain

Merrill Lutsky

CEO, Graphite

Matt Biilmann

CEO, Netlify

Ivan Burazin

CEO, Daytona

Paul Klein

CEO, Browserbase

Lance Martin

Researcher, Anthropic

Hunter Walk

Homebrew

Yohei Nakajima

Untapped Capital

Matthew Lenhard

Vercel

Orange Collective

And 90+ more angel investors from leading AI, developer tools, and infrastructure companies.

Thesis

Legacy observability platforms were built for 2010. Sazabi is what Datadog looks like if you built it in 2026 — with a purpose-built AI agent, its own storage layer, and a chat interface instead of dashboards.[1] [2] The longer-term bet: self-healing software — automating the second half of the software lifecycle the way Cursor and Claude Code automated the first.[8]
  1. 01

    The outer loop is the new bottleneck. AI coding tools have permanently raised baseline code output. The rate limiter has shifted to the second half of the software lifecycle: monitoring, debugging, and incident response. Teams are shipping faster than ever — and discovering bugs in production faster than ever.

  2. 02

    Datadog faces an innovator's dilemma it cannot escape. The platform, pricing model, and organizational structure are all optimized for dashboards and manual investigation. Rebuilding for AI would require cannibalizing their core business. They won't.

  3. 03

    Owning the full stack is the moat. Sazabi built its own storage layer, agent, and interfaces. This means it can memorize every incident your system has ever had, optimize query patterns specifically for AI access, and improve from the collective experience of all customers. Competitors sitting on top of Datadog cannot replicate this.

  4. 04

    Sherwood is purpose-built for this specific problem. 12 years in DevOps and observability — Crunchbase, Brex, 11x. Founded Brex's observability team. 2x YC founder with a successful exit. He has been building toward this exact product for his entire career.[7]

Sherwood Callaway — The End of Manual Debugging[8]

Problem

AI solved the first half of software engineering. The second half is still broken.

Teams using Claude Code or Cursor are shipping code 5–10× faster than they were two years ago. The bugs didn't slow down proportionally. If anything, they increased — more code reaching production faster means more surface area for things to go wrong.

The tool most engineering teams reach for when something breaks? Datadog — a company founded in 2010, three years before the first transformer paper. Its product is designed around manual dashboard investigation, complex query languages (PromQL, Datadog's own DSL), and niche expertise. It was an elegant solution for 2010. It feels excruciatingly painful next to Claude Code.

Right now, every engineering team in the world is coming to the same conclusion: it doesn't matter how fast you can write code if your users are constantly running into bugs and your app is constantly going down.[6]

5–10×

Code output increase

With modern AI coding tools

0

Code changes required

To set up Sazabi on 35+ platforms

<15 min

Time to value

From zero to live alerts

Why Now

The market already agrees. Legacy platforms can't adapt.

From Sazabi customers and angel investors — people who have lived inside the largest observability deployments in the world.

Having seen the product myself, I can vouch that it is miles ahead of any o11y product I've used and represents a foundational shift in how observability will be done in the future.

Matthew Lenhard

Matthew Lenhard[6]

Tech Lead, AI Gateway · Vercel

The outer loop is the new bottleneck, and software is more unstable than ever. Building on top of legacy platforms limits your ability to innovate. Today's teams demand a new set of opinionated, end-to-end AI tools to move at the speed of creation.

Merrill Lutsky

Merrill Lutsky[6]

CEO · Graphite

Can't escape the Sazabi! We've been on the product for a while now and it's awesome, proactively catching errors through our logs and creating fixes. We've merged many Sazabi PRs to Superset.

Kiet Ho

Kiet Ho[6]

CEO · Superset (P26)

Datadog faces a classic innovator's dilemma — and it's not going to win.

The revenue model. Datadog charges per host, per custom metric, per log ingested, per APM trace. Their enterprise sales motion assumes a dedicated observability team who configures dashboards and writes alert rules for months before getting value. An AI-first product that "just works" in 15 minutes would destroy this model.

The product architecture. Every AI feature Datadog has shipped is stateless — each investigation starts from scratch, with no memory of past incidents, no understanding of how your system has evolved. This isn't a feature gap. It's a fundamental architectural constraint. You can't bolt memory onto a stateless platform.

The organizational structure. Datadog has thousands of engineers building and maintaining dashboards, query languages, and integrations that a truly AI-native product wouldn't need. The company cannot unwind itself to compete.

These companies were built on a set of assumptions that no longer hold true. Everything about them, from their technology to their organizational hierarchies and business models, is structurally misaligned with AI.
— Sherwood Callaway, Founder & CEO of Sazabi[6]

How It Works

Three steps. Under 15 minutes. No code changes required.

Step 01

Connect GitHub

Sazabi gets read access to your codebase. It uses this to understand your architecture, correlate errors to specific code, and eventually open pull requests.

Step 02

Install the Slack app

Slack is a primary application entry point — not a notification channel. Alerts, investigations, and fixes all flow through Slack natively.

Step 03

Start sending logs

For 35+ services — Vercel, AWS, GCP, Supabase, Cloudflare, Neon, Temporal, and more — this requires zero code or config changes. The log intake endpoint is OpenTelemetry-compatible.

After setup, it runs itself.

Automatic system map. Sazabi scans your logs and codebase to register all key services, components, and product features. It generates a live status page for your application — any team member can see overall system health and any ongoing incidents in real time.

Intelligent alerting. Background agents monitor your logs, codebase, and infrastructure for anomalies — from unfamiliar errors to traffic spikes to crashed pods and failed deployments. When a legitimate issue is detected, Sazabi sends a rich Slack alert with root cause and suggested fix. Duplicate alerts are automatically suppressed — Sazabi checks every new alert against existing ones before it reaches you.

Multiplayer debugging. Sazabi conversations are shared by default. Your entire team can debug the same incident together in real time — the first truly multiplayer observability platform.[6]

Autonomous fixes. When you're ready to act, Sazabi can open a pull request against your repo. Sazabi recently outperformed Claude Code on TerminalBench 2 — not a benchmark they expected to lead, but a useful signal about the underlying agent quality.

Security & compliance — out of the box

Certifications
SOC 2 Type 2HIPAAGDPRISO 27001

Enterprise-grade certifications achieved in the company's first year of operation.

Logs Are All You Need

The most controversial idea behind the product.

Conventional wisdom says observability requires three pillars: logs, metrics, and traces. Sazabi disagrees.

No metrics. No traces. Only logs.

Why it's defensible. "Fundamentally, logs are just events, metrics are aggregated events, and traces are basically correlated events. We only accept logs, and we create metrics and traces from those logs on the back end."[10] Logs, metrics, and traces are not three different data types; they're three different views of the same underlying reality. Sazabi computes the views you need on demand.

Why it produces a better product. Sazabi's storage layer, query patterns, and agent are all optimized for a single data type. The architecture uses materialized views and LM-generated summaries of log windows — "We can take an hour's worth of log data and summarize that into a much smaller package using language models. You only have to query the summary."[10] Logs are also much easier to instrument than metrics or traces — for most teams, getting started means zero code changes.

Why AI makes it possible now. Three years ago, log analysis meant regex and pattern matching. Today, an LLM can read your log stream, understand root causes, correlate incidents across services, and explain exactly what happened in plain English. The "logs are all you need" thesis only holds because AI has transformed what you can do with raw log data.[6]

Lots of engineers are skeptical about this idea, but we're winning them over. If you have doubts about what AI can do with your log stream, I encourage you to try Sazabi and find out. You'll be very impressed.
— Sherwood Callaway[6]

Market

The highest-density version of their ICP is inside YC right now.

Sazabi's stated ICP: VC-backed tech startup, Seed to Series C, founded in the last three years, 5 to 50 engineers, moving fast with AI tools. This describes every company in the current YC batch — and every company in every future YC batch, indefinitely.

The ICP is also the fastest-growing segment in the tech economy. Every week, another cohort of AI-native startups gets funded and immediately hits the outer loop problem. They didn't grow up with Datadog and have no loyalty to it. They want something that works like Claude Code, not something that works like enterprise software from 2010.

Near term — AI-native startups

The YC cohort and similar AI-native teams. Fast decision cycles, single technical decision-maker, high willingness to pay for developer experience. Sazabi is already embedded in P26 through YC.

Long term — the full observability market

Gartner projects the observability platform market will reach $14.2B by 2028.[9] Datadog, Sentry, New Relic, Dynatrace, and Grafana together represent tens of billions in enterprise value. Each faces the same structural problem. Sazabi's pitch to enterprises comes later — but the addressable market is enormous.

Every YC company is a software company. Every software company has an observability solution. Sazabi should be that solution. I'll be following in a tradition of great YC companies — Brex, Deel, Rippling — that sold into their batchmates really successfully.
— Sherwood Callaway[8]

Competitive landscape

Four categories of competition. Sazabi is positioned against all of them.

Each competitor category has a structural limitation. Sazabi's vertical integration is the answer to all four.

AI SRE Startups

Layer on top

Traversal, Resolve, TierZero. These companies sit on top of Datadog and similar platforms. Makes sense for enterprises locked into multi-year contracts — but the ceiling is defined by the underlying platform. Sazabi's own storage unlocks capabilities these companies can never replicate.

DIY Solutions

In-house

Ramp's "On-Call Assistant" and similar internal tools built by smart engineering teams. These prove the demand — but it doesn't make sense for every company to independently build and maintain such a system. Sazabi builds the best version of this once and lets the entire industry benefit.

Legacy Players

Structurally misaligned

Datadog, Grafana, Sentry, Axiom. These companies have been adding AI features for years. None work well because they're stateless — each investigation starts fresh, with no memory of past incidents and no understanding of how your system has evolved. Can't be fixed without rebuilding from scratch.

LLM Observability

Narrow scope

Arize, Braintrust, Raindrop. These provide an important service for evaluating AI applications, but the scope is narrow — monitoring LLM quality, not the full production environment. Sazabi's thesis is that over time there will be no meaningful difference between observing agents and observing traditional software.[6]

The most important differentiator is our vertical integration. We're building every piece of the AI observability solution: the interfaces, the agent, the database. This allows us to do things that competitors fundamentally can't. We're the Apple of observability.
— Sherwood Callaway[6]

Founder deep dive

Sherwood's entire career was building toward this moment.

The origin. "I started my career in 2017 at Crunchbase as a junior engineer on the front end team. I somehow found myself working on our build and deploy system, which I rebuilt from the ground up. Crunchbase is where I became obsessed with the problem of software delivery."[6]

Brex — where observability became personal. "In February 2020, I joined Brex as the third infrastructure engineer. Brex leveled me up massively. I helped build all of the production infrastructure, build and deploy mechanisms and developer tools for the company during its period of hypergrowth. I also started their observability team."[6]

Opkit — the first company. "I left Brex in Summer 2021 to build my first company. Opkit was a YC-backed healthcare voice AI startup. I had a lot of fun building Opkit, but ultimately healthcare was not the right market for me. I found myself a lot more interested in the engineering challenges than the business ones. My co-founder and I decided to bow out. We sold the company to 11x."[7]

11x — the moment of clarity. "At 11x, I got the opportunity to build agents when they were still quite new. I also became obsessed with AI coding tools like Cursor and Claude Code. I was amazed by how we could suddenly code 10 times faster than before. But something else amazed me. It was the fact that when it came to fixing production, I was stuck using the same set of tools I had been using for my entire career: Datadog, Sentry, Grafana, and the rest. Compared with Cursor and Claude Code, these tools felt excruciatingly painful to use."[6]

On why this is his tombstone company. "I've been a DevOps infrastructure and observability engineer at early and growth-stage startups for my entire career. That's 12 years now. I know what it's like to build tools and systems for engineering orgs that are focused on product velocity above all else and just want to move faster. It didn't take long for me to realize that we were going to bring AI to observability and automate the second half of the software development life cycle. It also occurred to me that I was the perfect person to build this company."[6]

On doing YC a second time. "I feel way more confident as a founder today. I know how to fundraise, do a product launch, hire/fire people, etc. because I've done it all before. I've failed so many times and in so many different ways at this point. I'm not afraid of anything anymore."[6]

On the long-term vision. "Our mission is to become the default observability tool for all companies in what we call the AI era — starting with the year of agents in 2025 and forwards. Our vision is to create a world of self-healing software, where software improves itself without human intervention or direction. Cursor is automating the part of software development that's creating new features. A much bigger part of my job as a developer is maintaining software that's already been written. That's the opportunity for Sazabi."[8]

Founder & team

Sherwood Callaway

Sherwood Callaway

Repeat FounderExited

Founder & CEO

2x YC founder, a16z scout, and software engineer with 12 years of experience at top startups. Joined Brex in 2020 as its third infrastructure engineer, built production infrastructure through hypergrowth, and founded their observability team. Co-founded Opkit (YC S21), a healthcare voice AI startup, with Justin Ko — sold to 11x. At 11x, built production AI agents and became obsessed with the gap between AI coding tools and legacy observability. Left to build Sazabi.

Key team members

Justin Ko

Co-founder of Opkit alongside Sherwood. Worked together across 7 years and 4 companies. Early Brex.

Ed Carrel

Second infrastructure engineer at Brex. Early at Rootly, Looker, and InstaCart. True infrastructure expert.

Alex Holovach

Owned Observability Systems at a Fortune 500. Previously founded observability startup Kubiks.

Tom Nagengast

Early member of Replit's data engineering team. AI coding wizard.

Rupa Vemulapalli

Waterloo CS grad. SRE at Bell. Previously started an observability startup of her own.

Hadley Callaway

Chief of Staff. Columbia BA in CS. Previously an engineer at Brex and Doppel.

Risks & mitigations

Risk

"Logs are all you need" proves wrong — some enterprise workloads genuinely require distributed traces or custom metrics.

Mitigation

The OpenTelemetry-compatible intake means adding metric or trace support is an architectural extension, not a rewrite. Current alpha traction validates the core thesis at the target ICP. Sazabi is not initially targeting enterprises where this concern would first appear.

Risk

Greenfield enterprise sales is slow and political — replacing Datadog at large companies takes months of procurement.

Mitigation

Sazabi's stated ICP is Seed-to-Series C startups (5–50 engineers) where a single technical decision-maker can switch tools in a day. The YC cohort is the densest possible concentration of this buyer. Enterprise comes later, if at all.

Risk

Datadog or Sentry ships a competitive AI agent with institutional memory.

Mitigation

Sherwood's structural argument is compelling: Datadog's revenue model depends on dashboards, custom metrics, and per-host pricing. A truly AI-native product would cannibalize those lines. The architectural debt is also real — bolting memory onto a stateless platform is fundamentally different from building for it from day one.

Risk

Solo founder execution risk — Sherwood is carrying significant technical and organizational load.

Mitigation

Justin Ko (co-founder of Opkit, 7 years working with Sherwood) joins as a key technical hire. The team of nine includes domain experts in each critical area. Sherwood has proven he can recruit, lead, and ship under pressure across multiple companies.

What we're watching

  • First logo customers outside the YC cohort — particularly post-Series A companies and teams that have already tried Datadog.
  • GA launch and official seed round announcement — both expected in the near term.
  • "Logs are all you need" at scale — does the thesis hold for high-volume, latency-sensitive production environments?

References

  1. [1]Sazabi — YC Profile
  2. [2]Sazabi — Company Website
  3. [3]Sazabi — LinkedIn
  4. [4]Sazabi — X/Twitter
  5. [5]Sazabi — Product Overview (YouTube)
  6. [6]Sazabi — Launch on Bookface (YC internal, P26)
  7. [7]Sherwood Callaway — "Round Two" personal blog
  8. [8]Sherwood Callaway — Founder Interview (YouTube)
  9. [9]Gartner Observability Platform Market Forecast (via Network World)
  10. [10]Paul Gillin — "Startup Sazabi bets on logs and AI agents to replace traditional observability stacks" (SiliconAngle)